Elsevier

Geomorphology

Volume 243, 15 August 2015, Pages 40-50
Geomorphology

Estimation of debris flow triggering rainfall: Influence of rain gauge density and interpolation methods

https://doi.org/10.1016/j.geomorph.2015.04.028Get rights and content

Highlights

  • Rainfall estimation uncertainty leads to large errors in rainfall thresholds.

  • The nearest neighbor method is less biased for debris flow rainfall estimation.

  • Complexity of interpolation method has no apparent benefit on threshold estimation.

Abstract

The forecast of debris flow occurrence relies mainly on empirical rainfall intensity–duration thresholds, which are based on rain gauge observations. This work focuses on the effect of rainfall estimation uncertainty on the estimation of debris flow triggering rainfall events and on the identification of rainfall thresholds for debris flow occurrence. Specifically, the influence of rain gauge network density and the interpolation method on the estimation of debris flow triggering rainfall is investigated. These questions are examined using high-resolution, carefully corrected radar data to represent space–time patterns of true precipitation at the debris flow initiation points and in the surrounding area. Radar rainfall fields are sampled by simulated rain gauge networks, stochastically generated with varying rain gauge densities. Based on these networks, rainfall is estimated by using three rainfall interpolation methods: nearest neighbor (NN), inverse distance weighting (IDW) and ordinary kriging (OK). Results show that NN provides the estimates with bias smaller than IDW and OK but larger estimation variance. Overall, decrease in gauge density leads to increased underestimation and increased estimation variance of debris flow triggering rainfall. Rainfall estimation error leads to large underestimation of the intensity–duration thresholds. However, comparison of results shows that no particular benefit in intensity–duration threshold estimation is obtained by using approaches that are more complex than the NN method.

Introduction

Rainfall-induced landslides and debris flows pose a significant and widespread hazard, resulting in a large number of casualties and enormous economic damages worldwide (Salvati et al., 2010, Petley, 2012, Borga et al., 2014, Dowling and Santi, 2014, Melillo et al., 2014). Considerable research efforts have been made so far to determine the rainfall amount required for rainfall-induced landslides or debris flows (Salvati et al., 2010). Rainfall thresholds are often used to identify the local or regional rainfall conditions that, when reached or exceeded, are likely to result in landslides or debris flows (e.g., Caine, 1980, Wieczorek, 1996, Deganutti et al., 2000, Cannon and Gartner, 2005, Guzzetti et al., 2008, Frattini et al., 2009, Saito et al., 2010, Chen et al., 2011, Jakob et al., 2012). Following the pioneering works of Caine (1980) and Innes (1983), rainfall thresholds for shallow landslides and debris flows (collectively termed ‘debris flows’ hereinafter) were determined at the local, regional, and global scales. Guzzetti et al. (2008) proposed a review of the literature on rainfall thresholds for the possible initiation of debris flows. The identification of rainfall thresholds is affected by a number of uncertainties that limit the accuracy of debris flow forecasts and warnings (Guzzetti et al., 2008, Kirschbaum et al., 2012, Nikolopoulos et al., 2014). A specific source of uncertainty lies in the estimation of the rainfall amounts that are responsible for debris flows (Guzzetti et al., 2007, Nikolopoulos et al., 2014).

Rain gauge data are the typical source of information for the definition of the rainfall thresholds. This implies that, with the exception of the very rare cases when rain gauge data are available at the debris flow initiation site, rainfall data for debris flow events are estimated based on data from more or less remote neighboring stations. This rainfall estimation is difficult for two main reasons. First, debris flow triggering storms are often characterized by large rainfall gradients, possibly influenced by complex topography (Marra et al., 2014). Second, unknown precipitation must be estimated at points where it exceeds a threshold and it often forms a local peak (Marra et al., submitted for publication). As a result, findings from the literature on rainfall spatial estimation (see Masson and Frei, 2014 and references therein), where the whole rainfall distribution is sought, may not be completely relevant for the case of debris flow triggering storms.

Inspection of the literature shows that little attention has been dedicated to the problem of rainfall estimation for debris flow triggering storms (Chiang and Chang, 2009, Melillo et al., 2014). While the spatial variability of rainfall fields is often identified as a potential problem, researchers often overlook it in debris flow threshold modeling primarily because of the scarcity of rain gauges, especially in mountainous areas (Jakob and Weatherly, 2003, Guzzetti et al., 2004). Debris flow triggering rainfall is often estimated based on the concept of the reference gauges (e.g., Jakob and Weatherly, 2003, Aleotti, 2004, Godt et al., 2006, Brunetti et al., 2010, Berti et al., 2012, Schneuwly-Bollschweiler and Stoffel, 2012), which relies on the observation at the gauge nearest to the debris flow location. Nikolopoulos et al. (2014) have shown that inferring rainfall properties at initiation points from the closest rain gauge is associated with significant uncertainty. Furthermore, they showed that application of the closest gauge estimates leads to significantly underestimated thresholds of debris flow occurrence with a subsequent degradation of the efficiency of the warning procedures based on these thresholds. However, the work of Nikolopoulos et al. (2014) is based on a simulation exercise; hence, results may be claimed dependent on the realism of the simulation experiment itself.

The work presented in this paper has two main objectives. Firstly, we aim to quantify the dependence of the accuracy of debris flow-triggering rainfall estimation on rain gauge density and the use of various rainfall estimation procedures. Secondly, we evaluate the impact of the rainfall estimation errors on the identification of the intensity–duration (ID, hereinafter) thresholds used for predicting landslide and debris flow occurrence. These questions are examined by using high-resolution, carefully corrected radar rainfall data to represent space–time patterns of true precipitation at the debris flow initiation points and in the surrounding area. These rainfall fields are sampled by simulated rain gauge networks, stochastically generated with varying gauge densities. This approach is ideally suited to assess the properties of rainfall estimation errors in areas with strong gradients, where alternative methods are lacking. As such, this approach has been used in a number of studies for the assessment of the influence of rainfall sampling on areal rainfall estimation and rainfall–runoff modeling (Seed and Austin, 1990, Duncan et al., 1993, Fabry et al., 1994, Bradley et al., 2002, Volkmann et al., 2010). The present work is based on the availability of accurate radar-based estimates of rainfall fields for 10 storms, which triggered 82 debris flows in a mountainous region in the Eastern Italian Alps (the Upper Adige River basin).

Based on such simulated networks, four gauge densities are considered and three interpolation techniques are used to derive rainfall estimates at the debris flow initiation points. The rainfall estimation procedures considered here are: i) the nearest neighbor technique; ii) the inverse distance method; and iii) ordinary kriging. These estimates are then used to identify the ID thresholds for debris flows occurrence. Results are compared against the reference threshold obtained by using the actual rainfall data at debris flow initiation locations. In order to isolate the properties of the interpolation algorithms and to assess the effects of varying rain gauge density, we do not apply here further criteria for the reconstruction of rainfall events, i.e. application of minimum rainfall depth thresholds (Melillo et al., 2014, Nikolopoulos et al., 2014).

Section snippets

Study area and data

The area of study is part of the Upper Adige River basin in Northern Italy (Fig. 1), a mountainous region with more than 64% of the area located above 1500 m a.s.l., whereas only 4% of the area is located below 500 m a.s.l. The south-eastern sector of the region belongs to the Dolomites, the north-eastern part to the Noric Alps and the western sector to the eastern Rhaetian Alps, including the highest peak of Mt. Ortles (3902 m) (Norbiato et al., 2009a, Piacentini et al., 2012). The climate pattern

Methodology

To study issues related to rainfall estimation errors, the radar data were assumed to be the “true” observed rainfall and were therefore used as the reference for comparative analysis. An advantage of this is that variations in precipitation, as observed by radar and gauges, are represented empirically. In mountainous regions where orography influences precipitation patterns in complex ways (Molinié et al., 2012), the space–time patterns of non-stationary storm precipitation are very difficult

Analysis of the rainfall estimation error

For each realization of simulated rain gauge networks, rainfall estimates at DF locations were obtained from the three interpolation methods (NN, IDW, and OK). Interpolated values were then compared against the radar rainfall observations at the actual DF location, which serves as reference throughout the analysis. Error in rainfall estimation is expressed in terms of relative error, εr, defined asεr=RtotiDFRtotDFRtotDFwhere RtotiDF and RtotDF are interpolated and reference total event

Conclusions

The main objective of this work is to examine the problem of rain gauge-based estimation for rainfall events responsible of debris flows. We analyzed first the effect of rain gauge network density and spatial interpolation methods on the estimation of rainfall at the DF locations and then investigated the propagation of the rainfall estimation errors in the identification of ID thresholds. We have examined these issues by simulating rain gauge networks based on a random sampling exercise of

Acknowledgments

This work is supported by EU FP7 Marie Curie Actions IEF project PIEF-GA-2011-302720) (HYLAND, http://intra.tesaf.unipd.it/cms/hyland/). We sincerely acknowledge Ripartizione Opere Idrauliche, Autonomous Province of Bolzano (Italy), for providing access to the archive of debris flows and Ufficio Idrografico, Autonomous Province of Bolzano (Italy), for providing hydrological and meteorological data and for providing access to the weather radar information. We express our sincere gratitude to

References (67)

  • C.D. Lloyd

    Assessing the effect of integrating elevation data into the estimation of monthly precipitation in Great Britain

    J. Hydrol.

    (2005)
  • F. Marra et al.

    Radar rainfall estimation for the identification of debris-flow occurrence thresholds

    J. Hydrol.

    (2014)
  • E.I. Nikolopoulos et al.

    Impact of uncertainty in rainfall estimation on the identification of rainfall thresholds for debris flow occurrence

    Geomorphology

    (2014)
  • D. Norbiato et al.

    Controls on event runoff coefficients in the eastern Italian Alps

    J. Hydrol.

    (2009)
  • I. Pavlova et al.

    Debris flow activity related to recent climate conditions in the French Alps: a regional investigation

    Geomorphology

    (2014)
  • S. Peruccacci et al.

    Lithological and seasonal control on rainfall thresholds for the possible initiation of landslides in central Italy

    Geomorphology

    (2012)
  • D. Piacentini et al.

    Statistical analysis for assessing shallow-landslide susceptibility in South Tyrol (south-eastern Alps, Italy)

    Geomorphology

    (2012)
  • H. Saito et al.

    Relationship between the initiation of a shallow landslide and rainfall intensity—duration thresholds in Japan

    Geomorphology

    (2010)
  • A.W. Seed et al.

    Sampling errors for raingauge-derived mean areal daily and monthly rainfall

    J. Hydrol.

    (1990)
  • D.J. Seo

    Real-time estimation of rainfall fields using rain gage data under fractional coverage

    J. Hydrol.

    (1998)
  • M. Berti et al.

    Probabilistic rainfall thresholds for landslide occurrence using a Bayesian approach

    J. Geophys. Res.

    (2012)
  • A.A. Bradley et al.

    Raingauge network design using NEXRAD precipitation estimate

    JAWRA J. Am. Water Resour. Assoc.

    (2002)
  • M.T. Brunetti et al.

    Rainfall thresholds for the possible occurrence of landslides in Italy

    Nat. Hazards Earth Syst. Sci.

    (2010)
  • N. Caine

    The rainfall intensity: duration control of shallow landslides and debris flows

    Geogr. Ann. Ser. A. Phys. Geogr.

    (1980)
  • S. Cannon et al.

    Wildfire-related debris flow from a hazards perspective

  • J.C. Chen et al.

    Rainfall conditions for the initiation of debris flows during Typhoon Morakot in the Chen-Yu-Lan watershed in central Taiwan

  • G.J. Ciach et al.

    Radar–rain gauge comparisons under observational uncertainties

    J. Appl. Meteorol.

    (1999)
  • N.A. Cressie

    Statistics for Spatial Data

    (1991)
  • J. Creutin et al.

    Objective analyses and mapping techniques for rainfall fields: an objective comparison

    Water Resour. Res.

    (1982)
  • G. Crosta

    Regionalization of rainfall thresholds: an aid to landslide hazard evaluation

    Environ. Geol.

    (1998)
  • A.M. Deganutti et al.

    Rainfall and debris-flow occurrence in the Moscardo basin (Italian Alps)

  • C. Dowling et al.

    Debris flows and their toll on human life: a global analysis of debris-flow fatalities from 1950 to 2011

    Nat. Hazards

    (2014)
  • P. Frattini et al.

    Approaches for defining thresholds and return periods for rainfall-triggered shallow landslides

    Hydrol. Process.

    (2009)
  • Cited by (82)

    View all citing articles on Scopus
    View full text